Fix problems with losing tokens in the jobserver, reported by Grant
commit2fe13118167305240e227fb4e671c4aa8fa48f92
authorpsmith <psmith>
Tue, 3 May 2005 13:57:20 +0000 (3 13:57 +0000)
committerpsmith <psmith>
Tue, 3 May 2005 13:57:20 +0000 (3 13:57 +0000)
tree82f91dd78b555c7f6fe9f9c8a66bcb02d96db3f0
parent8f488ed50e0ac42fdce0b7cc3f03b58910973cd7
Fix problems with losing tokens in the jobserver, reported by Grant
Taylor.  There are two forms of this: first, it was possible to lose
tokens when using -j and -l at the same time, because waiting jobs were
not checked when determining whether any jobs were outstanding.  Second,
if you had an exported recursive variable that contained a $(shell ...)
function there is a possibility to lose tokens, since a token was taken
but the child list was not updated until after the shell function was
complete.

To resolve this I introduced a new variable that counted the number of
tokens we have obtained, rather than checking whether there were any
children on the list.  I also added some sanity checks to make sure we
weren't writing back too many or not enough tokens.  And, the master
make will drain the token pipe before exiting and compare the count of
tokens at the end to what was written there at the beginning.

Also:
  * Ensure a bug in the environment (missing "=") doesn't cause make to core.
  * Rename the .DEFAULT_TARGET variable to .DEFAULT_GOAL, to match the
    terminology in the documentation and other variables like MAKECMDGOALS.
  * Add documentation of the .DEFAULT_GOAL special variable.

Still need to document the secondary expansion stuff...
13 files changed:
ChangeLog
NEWS
doc/make.texi
filedef.h
job.c
job.h
main.c
read.c
tests/ChangeLog
tests/scripts/features/parallelism
tests/scripts/variables/DEFAULT_GOAL [moved from tests/scripts/variables/DEFAULT_TARGET with 57% similarity]
vmsfunctions.c
vmsjobs.c [new file with mode: 0644]